Impact of Radio Frequency Interference and Real-Time Spectral Kurtosis Mitigation

Abstract

We catalog the ubiquity of Radio Frequency Interference (RFI) plaguing every modern radio telescope and investigate several ways to mitigate it in order to create better science-ready data products for astronomers. There are a myriad of possible RFI sources, including satellite uplinks and downlinks, cellular communications, air traffic radar, and natural sources such as lightning. Real-time RFI mitigation strategies must take these RFI characteristics into account, as the interfering signals can look significantly different at very high time and frequency resolutions. We examine Spectral Kurtosis (SK) as a real-time statistical RFI detection method, and compare its flagging efficacy against simulated RFI witha wide range of signal characteristics. We found to be weak against signals with a 50% effective duty cycle, as well as low signal-to-noise ratio sidelobe spillover from strong and frequency-wide RFI. Coarsening the SK time resolution improved flagging, as did using multi-scale SK, which averages adjacent time-frequency pixels with small rolling windows to circumvent the weakness to 50% duty cycle signals. Multiscale SK raised flagging above 90% for almost all cases, and as long as the amount of channels included in the multi-scale window wasn’t wider than the RFI signal, there was no significant increase in false positive rate. Simulated realistic incoherent astronomical signals were not detected by SK at all, as expected. To simulate real-time SK RFI detection in a real data set, raw, unaveraged data was taken with the Robert C. Byrd Green Bank Telescope (GBT). The observation targets included one pulsar, two neutral hydrogen (HI) galaxies, the Milky Way HI emission, and a hydroxyl megamaser. These targets are all easily observable on short timescales but are also nearby several sources of RFI. Flagged data was replaced with representative Gaussian noise using the statistics of adjacent time-frequency pixels. We run different variations of SK detection on copies of the raw datasets and compared to the original, to see how well the RFI was removed and if the science data product was affected in any way. The spectral line targets are all completely ignored by SK , while the pulsar results decreased in quality due to the noise replacement averaging over the time variable structure, unless care was taken to flag data on timescales shorter than the pulse length. In these cases, single pulse signal-to-noise ratio was marginally improved.